Recent research on sepsis uses Deep Reinforcement Learning (DRL) to personalize treatments according to patients’ physiological characteristics, and thus render treatments more effective. However, current approaches rely on the relational data model, which struggles with representation of the complex relationships within medical data; thus, research on intelligent healthcare increasingly relies on knowledge graphs instead. Moreover, the output of these DRL approaches is a recommended action, i.e., a dosage; however, clinicians need the contextualization of such recommendations in order to decide whether to follow it.

We present a neuro-symbolic architecture for personalized sepsis treatments based on a graph-centric foundation. The architecture is based on representing medical data as a knowledge graph and learning via graph neural networks; their combination enables the inherent capturing of relationships and their native integration into reasoning, which may thus render recommendations more effective. Moreover, the architecture employs formally specified graph queries over the knowledge graph to contextualize personalized treatments, i.e., provide supporting information. We exemplify the architecture based on a widely-used medical dataset.